1. INTRODUCTION
Online hate speech has become a global concern, amplified by the rapid expansion of social
media platforms and the limited effectiveness of moderation systems, combined with a lack of
willingness even to moderate properly, as many social media platforms’ business models
equals large amounts of traffic, which you tend to achieve by promoting polarized views. In the
MENA region, online abuse often intersects with deeply rooted gender norms, political
polarisation, and identity-based discrimination. Women who participate in public debate—
journalists, activists, politicians, and young users—are particularly exposed to targeted
harassment, threats, and misogynistic narratives.6
Gendered and sexist hate speech frequently overlaps with other forms of discrimination,
including racism, and religious intolerance. This convergence exacerbates harm, discourages
participation, and undermines fundamental human rights, including freedom of expression and
access to information, and thereby democracy.
Civil society organisations, journalists and researchers play a critical role in documenting and
reporting online harm, supporting affected individuals, and advocating for policy change. They
need more leverage, not least in the shape of quantitative data, in order to catalyse systemic
changes. The scarcity of updated data and comprehensive studies also speak to the need of
further research, and not least quantitative studies. However, the sheer volume and velocity of
online content make manual monitoring insufficient. AI-powered tools offer new possibilities to
systematically detect patterns of abuse, generate evidence, activate protection mechanisms
and strengthen advocacy efforts.
This publication aims to:
Present an AI-powered research tool designed to detect online gendered hate speech
in Egyptian Arabic;
Assess the usability, effectiveness and limitations of AI tools for CSOs, journalists and
researchers;
Provide recommendations for civil society, researchers, policymakers, newsrooms and
other stakeholders when using AI-powered research tools.
JOSA has developed a similar AI-model for Jordanian Arabic (“Nuha”) in parallel to this
project. During the training phase of Nuha, JOSA observed online gender-based violence
(GBV) targeting women activists. Comments on social media platforms directed at women
activists sought to intimidate and discourage them from participating in politics and activism.
Misogynistic irony and sarcasm was employed to bypass automated content moderation.
Cases documented by JOSA indicated that women politicians and activists, including feminist
activists, were more subjected to hate speech online. Hate speech against women in politics in
Jordan included demeaning comments, accusations of importing and implementing so called
6 The International Center for Journalists’ “Big Data Case Study” examines the prolific campaign of online
violence against journalist Ghada Oueiss in the context of her lived experience of the abuse, her
journalistic work, and the socio-cultural and political conditions in which the abuse occurs,
ICFJ_BigData_Ghada Oueiss_Online Violence.pdf
2